Satellite field verification results
Keith Bouma-Gregson | CA State Waterboards
January 2, 2020
Introduction
The CA State Waterboards contracted with the San Francisco Estuary Institute (SFEI) to use satellite imagery to estimate cyanobacterial abundance. Algorithms to estimate cyanobacterial abundance in the surface waters of Lake Erie have been developed by Wynne et al. (2008) and Lunetta et al. (2015). Working with NOAA staff, SFEI applied these algorithms to waterbodies in California.
To better understand the performance of the satellite tool, field verification sampling was conducted in 2019. This document presents the initial results from the 2019 field verification effort.
Methods
Field sampling
Seven sampling events occurred at four different waterbodies from August 1 to October 8, 2019 (Table 1). Only Clear Lake had satellite pixels with positive CIcyano values on the sampling days. There had been blooms at San Pablo Reservoir and Lake San Antonio, but they had dissipated to non-detects on the satellite by the time we were able to sample them. Lake Almanor was chosen due to its higher elevation and previous questionable CIcyano values generated by MERIS imagery.
| Name | County | Sampling date | Surface area (km2) | Elevation (m) |
|---|---|---|---|---|
| Clear Lake | Lake | Aug-7, Aug-16, Oct-8 | 180 | 432 |
| Lake Almanor | Plumas | Aug-15 | 113 | 1373 |
| Lake San Antonio | Monterrey | Aug-1 | 23 | 245 |
| San Pablo Reservoir | Contra Costa | Aug-12 | 60 | 96 |
Field sampling occurred on the same day as the Sentinel-3a flyover. In a waterbody three pixels were selected for sampling. Within a pixel, three samples were collected, for a total of nine measurement sites (Fig. 1). At each sampling point, a Malvern Panalytical Fieldspec Handheld2 Pro radiometer was used to collect radiance data, which can be converted to a CIcyano equivalent value for comparison to the satellite data. Each reading involved collecting ten measurements at 1 nm wavelength resolution from 300-900 nms, which were then averaged into a single value per nm for the reading. Readings were taken on a calibrated 10% Spectralon reflectance plate, the water, and the sky. All readings were taken 40-45 degrees altitude and 115-130 degrees alzimuth from the sun. Triplicate plate, water, and sky readings were collected at each sampling site. Additionally, Secchi depth was measured, and depth integrated grab samples (1 m) for chlorophyll-a were collected at each sampling site. Chlorophyll-a samples were immediately chilled on ice in a cooler and filtered onto 0.7 micrometer (Whatman GF/C) filters back on shore. The filters were then wrapped in aluminum foil and frozen until fluorometric analysis.
Figure 1. Sampling maps of Lake San Antonio. Left panel) Entire lake showing the pixel locations for satellite imagery. Colors show the locations of the sampling pixels. Right panel) Zoom in of the sampling area showing the three different sampling sites within each of the three pixels. Numbers are the pixel ID for the satellite.
Field remote sensed reflectance calculations
The raw radiance values (W/m2/nm/sr) from the radiometer were converted to remote sensed reflectance (Rrs) values per nm/sr using the program test_asd_group.exe provided to the Waterboards by NOAA staff. The calculations take the plate, water, and sky readings and calculate Rrs based on equation 4 in Mobley 1999: \[
\begin{equation}
R_{rs} = \frac{L_{water} - \rho * L_{sky}}{L_{plate}}
\end{equation}
\] {#eq:1} Where L indicates the radiance values, and \(\rho\) represents the proportion of sky radiance that is reflected by the water’s surface.
Cyanobacterial index calculations
The Rrs values were used to calculate the cyanobacterial indices CI and CIcyano. The CI value is derived from the spectral shape (SS) at 681 nm and is calculated with equation 1 in Wynne et al. 2008: \[ \begin{equation} SS(681) = rrs681 - rrs665 - (rrs709 - rrs665) * \frac{681-665}{709-665} \end{equation} \] {#eq:2} More negative SS(681) values represent higher cyanobacterial abundances in the surface waters of a pixel. To transform it into positive values the SS(681) is converted to CI by: \[ \begin{equation} CI = -1*SS(681) \end{equation} \] {#eq:3} CI values <0 indicate no cyanobacteria present. However, certain water conditions can generate positive CI values, even when there are no cyanobacteria present. To reduce the frequency of false positives, an exclusionary criteria (Matthews et al. 2012 and Lunetta et al 2015) based on the spectral shape at 665 nm is used: \[ \begin{equation} SS(665) = rrs665 - rrs620 + (rrs620 - rrs681) * \frac{665-620}{681-620} \end{equation} \] {#eq:4} When SS(665) is >0 it indicates cyanobacteria present in the water and when it is <0, cyanobacteria is predicted to be absent. SS(665) is then transformed into an exclusion criteria value of 1 when SS(665) >0 and 0 when SS(665) <0. \[ \begin{equation} f(SS(665))=\begin{cases} 1, & \text{if SS(665) > 0}.\\ 0, & \text{if SS(665) < 0}. \end{cases} \end{equation} \] {#eq:5} The CI value and f(SS(665)) are combined to create a new indice called CIcyano which is calculated by: \[ \begin{equation} CI_{cyano} = f(SS(665)) * CI \end{equation} \] {#eq:6}
This will render all measurements with f(SS(665)) = 0 as non-detects, even if they had a positive CI value.
Satellite data calculations
The satellite CIcyano value is calculated from the OLCI sensor on Sentinel-3a satellite. The value is calculated by NOAA using top-of-atmosphere reflectance values applied to equations 1-5 above (Wynne et al. 2018). NOAA delivers the CIcyano product for each pixel as an integer ranging in values 0-250, corresponding to increasing CIcyano value. The pixel integer value is converted to CIcyano with: \[ \begin{equation} CI_{cyano} = 10^{0.012 * PixInteger - 4.2} \end{equation} \] {#eq:7}
CIcyano has a range of 0.000063 - 0.063270. To make these values easier to work with, SFEI and the CA Waterboards then multiply CIcyano by a constant to make the values easier to work with by putting them on a scale of 1-1000. This index is called CImod and calculated by: \[ \begin{equation} CI_{mod} = CI_{cyano} * 15805.18 \end{equation} \] {#eq:8}
Results
Lake conditions
Lakes ranged in phytoplankton concentrations, with median chl-a concentrations ranging from 1-40 ug/L (Fig. 2) and Secchi depths ranged from <1 - 4.5 meters (Fig. 3). Only in Clear Lake were cyanobacterial colonies visible (Fig. 4), however, microscopic analysis of samples did identify cyanobacterial taxa in some waterbodies, including Dolichospermum, Microcystis, and Gloeotrichia (Table 2). No cell counts of cyanobacteria abundance were performed.
Figure 2. Chlorophyll-a concentrations at sampling locations in waterbodies. (Note: data was not collected for Clear Lake on Aug-16 and Oct-8).
Figure 3. Secchi depth at sampling locations in waterbodies. (Note: data was not collected for Clear Lake on Aug-16 and Oct-8).
| Name | Sampling_date | Cyanobacteria |
|---|---|---|
| Clear Lake | Aug-7 | Microcystis, Gloeotrichia, Dolichospermum |
| Clear Lake | Aug-16 | No data |
| Clear Lake | Oct-8 | No data |
| Lake Almanor | Aug-15 | None |
| Lake San Antonio | Aug-1 | Dolichospermum |
| San Pablo Reservoir | Aug-12 | Dolichospermum |
Figure 4. Photographs of water conditions on sampling days. a) Lake San Antonio, b) Lake Almanor, c) Clear Lake Aug-16, and d) Clear Lake Oct-08.
Remote sensed reflectance spectra
The field collected remote sensed reflectance (Rrs) data showed chlorophyll-a absorbption decrease ~681 nm in samples with high chlorophyll-a concentrations, Clear Lake and Lake San Antonio (Fig. 5 top row). Lake Alanor had the lowest chl-a concentratios (Fig. 2) and little chlorophyll a absorption an no phycocyanin absorption. San Pablo Reservoir demonstrated lower chlorophyll absorption. The Lake Almanor and San Pablo Reservoir CImod values were below satellite detection levels due to an absence of the phycocyanin absorption at 620 nm.
